> On Apr 17, 2018, at 21:46, Luis Henriques <lhenriques@xxxxxxxx> wrote: > > Hi! > > I see the following commit is queued for the kernel cephfs client: > > commit e2311baa73a883eb8501c895cc817d3c96c3b896 > Author: Yan, Zheng <zyan@xxxxxxxxxx> > Date: Sun Apr 8 09:54:39 2018 +0800 > > ceph: check if mds create snaprealm when setting quota > > If mds does not, return -EOPNOTSUPP. > > [ This is related with http://tracker.ceph.com/issues/23491 ] > > However, I'm not sure it is doing the right thing (although I confess > I'm not sure either what the right thing would be...) > > This commit is doing 2 things: > > 1) it returns an error if a user tries to set quotas and the MDS doesn't > support it (using the new snaprealm method) > > 2) it forbids reading the ceph.quota.* xattrs, even if they're set > > Regarding 1), it's a change in behaviour from previous kernels -- older > kernels allow these attributes to be set (but not read!), even if they > don't really support quotas. Also, the error that is returned > (-EOPNOTSUPP) is a bit misleading as the xattr is actually set in the > directory. any better error code? > > Regarding 2), it's a bit artificial to forbid the user to get these > attributes as we do actually have them available. If attributes have no effect, why can’t we treat them as nonexistence. > > I understand the rational behind this patch, but since quotas won't be > enforced anyway in this scenario, is there really any value added with > this patch? > > I would suggest 2 alternatives: > > 1. check cluster version and return the error in setxattr if version < > Mimic (but maybe still allow reading the xattrs?) > The problem is we run out of feature bits. No straight way to do the check > 2. do nothing -- if we're using a cluster that does not support quotas > anyway, maybe there's no point in having these different behaviours > for old and new clients. > This is confusing. we reclaim that > 4.16 kernel supports quota. I don’t think it’s a good that user successfully set quota on > 4.16 kernel, but find that quota has no effect. 4.16 kernel is the first one that support quota, does behavior change really matter? Regards Yan, Zheng > Thoughts? > > Cheers, > -- > Luis -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
